In [1]:
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.impute import SimpleImputer
import seaborn as sns
import numpy as np
import ydata_profiling
from sklearn.model_selection import train_test_split
from sklearn.pipeline import Pipeline
from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.linear_model import LogisticRegression
Upgrade to ydata-sdk

Improve your data and profiling with ydata-sdk, featuring data quality scoring, redundancy detection, outlier identification, text validation, and synthetic data generation.

In [2]:
df = pd.read_csv("Delinquency_prediction_dataset.csv")
df.head(10)
Out[2]:
Customer_ID Age Income Credit_Score Credit_Utilization Missed_Payments Delinquent_Account Loan_Balance Debt_to_Income_Ratio Employment_Status Account_Tenure Credit_Card_Type Location Month_1 Month_2 Month_3 Month_4 Month_5 Month_6
0 CUST0001 56 165580.0 398.0 0.390502 3 0 16310.0 0.317396 EMP 18 Student Los Angeles Late Late Missed Late Missed Late
1 CUST0002 69 100999.0 493.0 0.312444 6 1 17401.0 0.196093 Self-employed 0 Standard Phoenix Missed Missed Late Missed On-time On-time
2 CUST0003 46 188416.0 500.0 0.359930 0 0 13761.0 0.301655 Self-employed 1 Platinum Chicago Missed Late Late On-time Missed Late
3 CUST0004 32 101672.0 413.0 0.371400 3 0 88778.0 0.264794 Unemployed 15 Platinum Phoenix Late Missed Late Missed Late Late
4 CUST0005 60 38524.0 487.0 0.234716 2 0 13316.0 0.510583 Self-employed 11 Standard Phoenix Missed On-time Missed Late Late Late
5 CUST0006 25 84042.0 700.0 0.650540 6 0 48361.0 0.260688 Unemployed 7 Gold New York On-time Late Missed Missed Missed Late
6 CUST0007 38 35056.0 354.0 0.390581 3 0 4638.0 0.484265 employed 17 Platinum New York On-time Missed Missed Late Missed Late
7 CUST0008 56 123215.0 415.0 0.532715 5 0 55776.0 0.358695 EMP 1 Student New York On-time On-time On-time Late Missed Late
8 CUST0009 36 66991.0 405.0 0.413035 5 1 NaN 0.219854 Employed 12 Student Phoenix On-time On-time On-time Missed Late On-time
9 CUST0010 40 34870.0 679.0 0.361824 4 0 93922.0 0.333081 EMP 5 Business Los Angeles On-time Missed Missed On-time Missed Missed
In [3]:
from ydata_profiling import ProfileReport

# create the report
profile = ProfileReport(df, title="EDA Report")
profile.to_file("eda_report.html")

Summarize dataset:   0%|          | 0/5 [00:00<?, ?it/s]
  0%|                                                                                           | 0/19 [00:00<?, ?it/s]
 26%|█████████████████████▊                                                             | 5/19 [00:00<00:00, 36.86it/s]
 53%|███████████████████████████████████████████▏                                      | 10/19 [00:00<00:00, 40.94it/s]
100%|██████████████████████████████████████████████████████████████████████████████████| 19/19 [00:00<00:00, 32.85it/s]

Generate report structure:   0%|          | 0/1 [00:00<?, ?it/s]
Render HTML:   0%|          | 0/1 [00:00<?, ?it/s]
Export report to file:   0%|          | 0/1 [00:00<?, ?it/s]
In [4]:
profile.to_notebook_iframe()

Income has 39 (7.8%) missing values Missing Loan_Balance has 29 (5.8%) missing values Missing Customer_ID has unique values Unique Missed_Payments has 77 (15.4%) zeros Zeros Account_Tenure has 28 (5.6%) zeros Zeros

In [5]:
#changing EMP, employed  labes to Employed

df['Employment_Status'].replace({
    'employed': 'Employed',
    'EMP': 'Employed'
},inplace=True)
In [6]:
df['Employment_Status'].value_counts()
Out[6]:
Employment_Status
Employed         240
Unemployed        93
retired           87
Self-employed     80
Name: count, dtype: int64
In [7]:
# Number of records (rows, columns)
df.shape
Out[7]:
(500, 19)
In [8]:
# Column names and data types
df.dtypes
Out[8]:
Customer_ID              object
Age                       int64
Income                  float64
Credit_Score            float64
Credit_Utilization      float64
Missed_Payments           int64
Delinquent_Account        int64
Loan_Balance            float64
Debt_to_Income_Ratio    float64
Employment_Status        object
Account_Tenure            int64
Credit_Card_Type         object
Location                 object
Month_1                  object
Month_2                  object
Month_3                  object
Month_4                  object
Month_5                  object
Month_6                  object
dtype: object
In [9]:
# Check for missing values
df.isnull().sum()
Out[9]:
Customer_ID              0
Age                      0
Income                  39
Credit_Score             2
Credit_Utilization       0
Missed_Payments          0
Delinquent_Account       0
Loan_Balance            29
Debt_to_Income_Ratio     0
Employment_Status        0
Account_Tenure           0
Credit_Card_Type         0
Location                 0
Month_1                  0
Month_2                  0
Month_3                  0
Month_4                  0
Month_5                  0
Month_6                  0
dtype: int64
In [10]:
# Percentage of missing values per column
df.isnull().mean() * 100
Out[10]:
Customer_ID             0.0
Age                     0.0
Income                  7.8
Credit_Score            0.4
Credit_Utilization      0.0
Missed_Payments         0.0
Delinquent_Account      0.0
Loan_Balance            5.8
Debt_to_Income_Ratio    0.0
Employment_Status       0.0
Account_Tenure          0.0
Credit_Card_Type        0.0
Location                0.0
Month_1                 0.0
Month_2                 0.0
Month_3                 0.0
Month_4                 0.0
Month_5                 0.0
Month_6                 0.0
dtype: float64
In [11]:
# Check for duplicate rows
df.duplicated().sum()
Out[11]:
0

Data imputation income and loan balance missing values replaced with median credit score missing values replaced with mean

In [12]:
# Impute 'Income' and 'Loan_Balance' with median
median_imputer = SimpleImputer(strategy="median")
df[['Income', 'Loan_Balance']] = median_imputer.fit_transform(df[['Income', 'Loan_Balance']])

# Impute 'Credit_Score' with mean
mean_imputer = SimpleImputer(strategy="mean")
df[['Credit_Score']] = mean_imputer.fit_transform(df[['Credit_Score']])

# Confirm no missing values remain
print(df.isnull().sum())

Customer_ID             0
Age                     0
Income                  0
Credit_Score            0
Credit_Utilization      0
Missed_Payments         0
Delinquent_Account      0
Loan_Balance            0
Debt_to_Income_Ratio    0
Employment_Status       0
Account_Tenure          0
Credit_Card_Type        0
Location                0
Month_1                 0
Month_2                 0
Month_3                 0
Month_4                 0
Month_5                 0
Month_6                 0
dtype: int64
In [13]:
#  Keep only numeric columns 
numeric_df = df.select_dtypes(include=['int64', 'float64'])

# Correlation matrix
corr = numeric_df.corr()
In [14]:
%matplotlib inline
In [15]:
# Correlation matrix (numerical features only)
corr = numeric_df.corr()

# Heatmap of correlations
plt.figure(figsize=(10,6))
sns.heatmap(corr, annot=True, fmt=".2f", cmap="coolwarm")
plt.title("Correlation Matrix")
plt.show()
No description has been provided for this image
In [16]:
 #correlation of each variable with delinquency
corr_with_target = corr['Delinquent_Account'].sort_values(ascending=False)
print(corr_with_target)

Delinquent_Account      1.000000
Income                  0.043991
Credit_Score            0.034820
Debt_to_Income_Ratio    0.034386
Credit_Utilization      0.034224
Age                     0.022508
Loan_Balance           -0.005438
Missed_Payments        -0.026478
Account_Tenure         -0.039829
Name: Delinquent_Account, dtype: float64
In [17]:
# Pairplot for key risk factors
sns.pairplot(df[['Credit_Score', 'Income', 'Loan_Balance', 
                       'Credit_Utilization', 'Debt_to_Income_Ratio', 
                       'Delinquent_Account']], hue="Delinquent_Account")
plt.show()
No description has been provided for this image
In [18]:
#calculate and display delinquency rates (%) across different categories

categorical_cols = ['Employment_Status', 'Debt_to_Income_Ratio', 'Credit_Score' , 'Credit_Utilization' ]

for col in categorical_cols:
    delinquency_rate = df.groupby(col)['Delinquent_Account'].mean().sort_values(ascending=False)*100
    print(f"\nDelinquency Rate by {col}:\n", delinquency_rate)

  

Delinquency Rate by Employment_Status:
 Employment_Status
Unemployed       19.354839
Employed         16.250000
Self-employed    16.250000
retired          11.494253
Name: Delinquent_Account, dtype: float64

Delinquency Rate by Debt_to_Income_Ratio:
 Debt_to_Income_Ratio
0.313142    100.0
0.438178    100.0
0.243359    100.0
0.367570    100.0
0.366179    100.0
            ...  
0.269624      0.0
0.269559      0.0
0.269109      0.0
0.268534      0.0
0.552956      0.0
Name: Delinquent_Account, Length: 487, dtype: float64

Delinquency Rate by Credit_Score:
 Credit_Score
378.0    100.0
412.0    100.0
445.0    100.0
805.0    100.0
731.0    100.0
         ...  
526.0      0.0
528.0      0.0
534.0      0.0
535.0      0.0
567.0      0.0
Name: Delinquent_Account, Length: 235, dtype: float64

Delinquency Rate by Credit_Utilization:
 Credit_Utilization
0.448492    100.0
0.257505    100.0
0.427107    100.0
0.337007    100.0
0.575592    100.0
            ...  
0.401246      0.0
0.400141      0.0
0.398533      0.0
0.397710      0.0
1.025843      0.0
Name: Delinquent_Account, Length: 492, dtype: float64
In [19]:
# Employment Status
delinq_by_emp = df.groupby("Employment_Status")["Delinquent_Account"].mean() * 100

# Debt-to-Income Ratio (binning)
df["DTI_bin"] = pd.cut(df["Debt_to_Income_Ratio"], bins=[0,0.2,0.4,0.6,1.0])
delinq_by_dti = df.groupby("DTI_bin")["Delinquent_Account"].mean() * 100

# Credit Score (binning)
df["Credit_bin"] = pd.cut(df["Credit_Score"], bins=[300,500,600,700,850])
delinq_by_credit = df.groupby("Credit_bin")["Delinquent_Account"].mean() * 100

# Credit Utilization (binning)
df["Util_bin"] = pd.cut(df["Credit_Utilization"], bins=[0,0.3,0.5,0.7,1.0])
delinq_by_util = df.groupby("Util_bin")["Delinquent_Account"].mean() * 100

# Payment History (Missed Payments binning)
df["PayHist_bin"] = pd.cut(df["Missed_Payments"], bins=[0,2,5,10])
delinq_by_payhist = df.groupby("PayHist_bin")["Delinquent_Account"].mean() * 100
In [20]:
fig, axes = plt.subplots(3, 2, figsize=(12,10))

# Employment Status
delinq_by_emp.plot(kind="bar", ax=axes[0,0], color="skyblue", edgecolor="black")
axes[0,0].set_title("Delinquency Rate by Employment Status (%)")
axes[0,0].set_ylabel("Delinquency Rate (%)")

# Debt-to-Income Ratio
delinq_by_dti.plot(kind="bar", ax=axes[0,1], color="salmon", edgecolor="black")
axes[0,1].set_title("Delinquency Rate by DTI (Binned)")
axes[0,1].set_ylabel("Delinquency Rate (%)")

# Credit Score
delinq_by_credit.plot(kind="bar", ax=axes[1,0], color="lightgreen", edgecolor="black")
axes[1,0].set_title("Delinquency Rate by Credit Score (Binned)")
axes[1,0].set_ylabel("Delinquency Rate (%)")

# Credit Utilization
delinq_by_util.plot(kind="bar", ax=axes[1,1], color="yellow", edgecolor="black")
axes[1,1].set_title("Delinquency Rate by Credit Utilization (Binned)")
axes[1,1].set_ylabel("Delinquency Rate (%)")

# Payment History
delinq_by_payhist.plot(kind="bar", ax=axes[2,0], color="black", edgecolor="black")
axes[2,0].set_title("Delinquency Rate by Payment History (Binned)")
axes[2,0].set_ylabel("Delinquency Rate (%)")

# Hide empty subplot (bottom-right)
axes[2,1].axis("off")

plt.tight_layout()
plt.show()
No description has been provided for this image
In [21]:
df.head(10)
Out[21]:
Customer_ID Age Income Credit_Score Credit_Utilization Missed_Payments Delinquent_Account Loan_Balance Debt_to_Income_Ratio Employment_Status ... Month_1 Month_2 Month_3 Month_4 Month_5 Month_6 DTI_bin Credit_bin Util_bin PayHist_bin
0 CUST0001 56 165580.0 398.0 0.390502 3 0 16310.0 0.317396 Employed ... Late Late Missed Late Missed Late (0.2, 0.4] (300, 500] (0.3, 0.5] (2.0, 5.0]
1 CUST0002 69 100999.0 493.0 0.312444 6 1 17401.0 0.196093 Self-employed ... Missed Missed Late Missed On-time On-time (0.0, 0.2] (300, 500] (0.3, 0.5] (5.0, 10.0]
2 CUST0003 46 188416.0 500.0 0.359930 0 0 13761.0 0.301655 Self-employed ... Missed Late Late On-time Missed Late (0.2, 0.4] (300, 500] (0.3, 0.5] NaN
3 CUST0004 32 101672.0 413.0 0.371400 3 0 88778.0 0.264794 Unemployed ... Late Missed Late Missed Late Late (0.2, 0.4] (300, 500] (0.3, 0.5] (2.0, 5.0]
4 CUST0005 60 38524.0 487.0 0.234716 2 0 13316.0 0.510583 Self-employed ... Missed On-time Missed Late Late Late (0.4, 0.6] (300, 500] (0.0, 0.3] (0.0, 2.0]
5 CUST0006 25 84042.0 700.0 0.650540 6 0 48361.0 0.260688 Unemployed ... On-time Late Missed Missed Missed Late (0.2, 0.4] (600, 700] (0.5, 0.7] (5.0, 10.0]
6 CUST0007 38 35056.0 354.0 0.390581 3 0 4638.0 0.484265 Employed ... On-time Missed Missed Late Missed Late (0.4, 0.6] (300, 500] (0.3, 0.5] (2.0, 5.0]
7 CUST0008 56 123215.0 415.0 0.532715 5 0 55776.0 0.358695 Employed ... On-time On-time On-time Late Missed Late (0.2, 0.4] (300, 500] (0.5, 0.7] (2.0, 5.0]
8 CUST0009 36 66991.0 405.0 0.413035 5 1 45776.0 0.219854 Employed ... On-time On-time On-time Missed Late On-time (0.2, 0.4] (300, 500] (0.3, 0.5] (2.0, 5.0]
9 CUST0010 40 34870.0 679.0 0.361824 4 0 93922.0 0.333081 Employed ... On-time Missed Missed On-time Missed Missed (0.2, 0.4] (600, 700] (0.3, 0.5] (2.0, 5.0]

10 rows × 23 columns

In [ ]:
!pip install pycaret

Modeling using Gen AI ( pyncret )

In [23]:
# Import classification module
from pycaret.classification import *
In [24]:
from pycaret.classification import setup, compare_models, evaluate_model, predict_model

# Setup classification experiment
exp = setup(
    data = df,
    target = "Delinquent_Account",   # target variable
    session_id = 123,                # reproducibility
    train_size = 0.8,                # 80/20 split
    normalize = True,                # scale numeric features
    categorical_imputation = 'mode', # fill missing categorical values
    numeric_imputation = 'mean',     # fill missing numeric values
    verbose = False                  # suppress setup logs if you want a clean output
)
In [25]:
# Compare models (AutoML)
best_model = compare_models(sort = 'AUC')  # sorts models by ROC AUC
  Model Accuracy AUC Recall Prec. F1 Kappa MCC TT (Sec)
svm SVM - Linear Kernel 0.8400 0.5323 0.0000 0.0000 0.0000 0.0000 0.0000 0.3240
knn K Neighbors Classifier 0.8350 0.5303 0.0000 0.0000 0.0000 -0.0092 -0.0147 0.2510
et Extra Trees Classifier 0.8400 0.5159 0.0000 0.0000 0.0000 0.0000 0.0000 0.4630
nb Naive Bayes 0.8400 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.2580
dt Decision Tree Classifier 0.8400 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.2740
ada Ada Boost Classifier 0.8400 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.2840
gbc Gradient Boosting Classifier 0.8400 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.7340
lda Linear Discriminant Analysis 0.8400 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.2960
dummy Dummy Classifier 0.8400 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.2460
qda Quadratic Discriminant Analysis 0.8400 0.4942 0.0000 0.0000 0.0000 0.0000 0.0000 0.2750
ridge Ridge Classifier 0.8400 0.4796 0.0000 0.0000 0.0000 0.0000 0.0000 0.2720
lr Logistic Regression 0.8400 0.4790 0.0000 0.0000 0.0000 0.0000 0.0000 1.8750
rf Random Forest Classifier 0.8400 0.4710 0.0000 0.0000 0.0000 0.0000 0.0000 0.3880
In [26]:
#  Evaluate model using  ROC, PR curve, confusion matrix
evaluate_model(best_model)

interactive(children=(ToggleButtons(description='Plot Type:', icons=('',), options=(('Pipeline Plot', 'pipelin…

Interpreting model with SHAP. SHAP (SHapley Additive exPlanations) is a method to explain the predictions of a machine learning model by showing how much each feature contributes

In [27]:
# Only include tree-based models
best_model = compare_models(include=["lightgbm", "rf", "et", "dt"], sort="AUC")

# Now interpret works
interpret_model(best_model, plot="summary")
  Model Accuracy AUC Recall Prec. F1 Kappa MCC TT (Sec)
et Extra Trees Classifier 0.8400 0.5159 0.0000 0.0000 0.0000 0.0000 0.0000 0.4720
dt Decision Tree Classifier 0.8400 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.3160
rf Random Forest Classifier 0.8400 0.4710 0.0000 0.0000 0.0000 0.0000 0.0000 0.4820
No description has been provided for this image

SHAP summary plot shows the overall impact of features on a machine learning model's output (which features are most influential and how they affect the predictions with the most impactful feature at the top)

High Impact Features (at the top) → Customer_ID, Month_6_Late. Wide spreads mean they can shift predictions strongly in either direction.

Customer_ID → unusually important; this suggests data leakage since IDs shouldn’t affect outcomes. Needs investigation.

Month_6_Late → late in month 6 (red) raises default risk; not late (blue) lowers it.

Month_1_On-time → being on time lowers default risk; not on time raises it.

Location_Los Angeles → being in LA reduces risk; not in LA increases it.

Extra Trees, Decision Tree, Random Forest are showing high accuracy (0.84) but zero recall, precision, and F1 which signifies Severe class imbalance

Most customers in the dataset are not delinquent and models are just predicting “Not Delinquent” for everyone, which gives high accuracy but fails to detect delinquents (hence recall = 0).

AUC is near random (~0.5)Confirms the model isn’t separating delinquent(Customers who are paying their credit obligations on time) vs. non-delinquent customers (Customers who have missed payments).

ACTION: Use Smote to solve the imbalance

In [28]:
#enable  balancing 

from imblearn.over_sampling import SMOTE

exp = setup(
    data=df,
    target="Delinquent_Account",
    session_id=123,
    train_size=0.8,
    normalize=True,
    categorical_imputation="mode",
    numeric_imputation="mean",
    fix_imbalance=True,
    fix_imbalance_method=SMOTE()   
)
  Description Value
0 Session id 123
1 Target Delinquent_Account
2 Target type Binary
3 Original data shape (500, 23)
4 Transformed data shape (772, 56)
5 Transformed train set shape (672, 56)
6 Transformed test set shape (100, 56)
7 Numeric features 8
8 Categorical features 14
9 Rows with missing values 16.2%
10 Preprocess True
11 Imputation type simple
12 Numeric imputation mean
13 Categorical imputation mode
14 Maximum one-hot encoding 25
15 Encoding method None
16 Fix imbalance True
17 Fix imbalance method SMOTE(k_neighbors=5, random_state=None, sampling_strategy='auto')
18 Normalize True
19 Normalize method zscore
20 Fold Generator StratifiedKFold
21 Fold Number 10
22 CPU Jobs -1
23 Use GPU False
24 Log Experiment False
25 Experiment Name clf-default-name
26 USI bd4a

Summary

Original data shape → (500, 23) Transformed data shape → (772, 56), SMOTE created synthetic samples of the minority class (delinquent customers/training data) to balance the dataset.

Train set (672, 56) / Test set (100, 56).Data was split into 80% train / 20% test after balancing.

In [29]:
#  Compare models after smote
best_model = compare_models(sort="Recall")
  Model Accuracy AUC Recall Prec. F1 Kappa MCC TT (Sec)
knn K Neighbors Classifier 0.3100 0.4568 0.6690 0.1435 0.2362 -0.0364 -0.0725 0.2660
lda Linear Discriminant Analysis 0.5525 0.4398 0.3738 0.1422 0.2014 -0.0299 -0.0309 0.2640
svm SVM - Linear Kernel 0.8375 0.5213 0.0167 0.0333 0.0222 0.0136 0.0146 0.3130
lr Logistic Regression 0.8400 0.4506 0.0000 0.0000 0.0000 0.0000 0.0000 0.2620
nb Naive Bayes 0.8400 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.2970
dt Decision Tree Classifier 0.8400 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.2850
ridge Ridge Classifier 0.8400 0.4622 0.0000 0.0000 0.0000 0.0000 0.0000 0.2970
rf Random Forest Classifier 0.8400 0.4370 0.0000 0.0000 0.0000 0.0000 0.0000 0.4660
qda Quadratic Discriminant Analysis 0.8400 0.4917 0.0000 0.0000 0.0000 0.0000 0.0000 0.2960
ada Ada Boost Classifier 0.8400 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.3170
gbc Gradient Boosting Classifier 0.8400 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.4090
et Extra Trees Classifier 0.8400 0.5674 0.0000 0.0000 0.0000 0.0000 0.0000 0.4670
dummy Dummy Classifier 0.8400 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.2870
In [30]:
# Evaluate the best model
evaluate_model(best_model)

interactive(children=(ToggleButtons(description='Plot Type:', icons=('',), options=(('Pipeline Plot', 'pipelin…

this plot provides a visual summary of all the data preprocessing and modeling steps that PyCaret performed behind the scenes to create the final model

Raw data: This is your initial, unprocessed dataset.

SimpleImputer(one for numerical, one for categorical): ensures data is clean before it's used for training.

OneHotEncoder: converts categorical data into a numerical format createing new binary columns that the machine learning model can understand.

TargetEncoder: replaces each category with a numerical value based on the mean of the target variable for that category.

FixImbalance: addresses the class imbalance in the target variable (Delinquent_Account)

StandardScaler: scales numerical features so they have a mean of 0 and a standard deviation of 1.

KNeighborsClassifier: where the model is trained on the preprocessed data.

Hyperparameter Tuning

calibration ( PyCaret, calibrate_model()) : improve the probability estimates of a trained classifier by applying probability calibration.Helps improve KNN model predictive performance , especially the k value.

In [31]:
# Create KNN model
knn = create_model('knn')

# Calibrate the KNN model
calibrated_knn = calibrate_model(knn, method='isotonic') 

# Plot calibration curve
plot_model(calibrated_knn, plot='calibration')
  Accuracy AUC Recall Prec. F1 Kappa MCC
Fold              
0 0.3500 0.4069 0.5000 0.1154 0.1875 -0.0744 -0.1321
1 0.3500 0.4755 0.6667 0.1429 0.2353 -0.0156 -0.0306
2 0.2500 0.1912 0.5000 0.1000 0.1667 -0.1111 -0.2425
3 0.3500 0.6544 0.6667 0.1429 0.2353 -0.0156 -0.0306
4 0.4000 0.5539 0.6667 0.1538 0.2500 0.0083 0.0147
5 0.3750 0.4608 0.6667 0.1481 0.2424 -0.0040 -0.0075
6 0.2750 0.4740 0.5714 0.1333 0.2162 -0.0943 -0.1899
7 0.2500 0.3290 0.4286 0.1034 0.1667 -0.1605 -0.3058
8 0.3000 0.3377 0.5714 0.1379 0.2222 -0.0832 -0.1584
9 0.4500 0.5736 0.8571 0.2222 0.3529 0.1039 0.1791
Mean 0.3350 0.4457 0.6095 0.1400 0.2275 -0.0447 -0.0904
Std 0.0624 0.1285 0.1158 0.0326 0.0507 0.0713 0.1355
  Accuracy AUC Recall Prec. F1 Kappa MCC
Fold              
0 0.7250 0.4265 0.0000 0.0000 0.0000 -0.1579 -0.1588
1 0.7000 0.4853 0.0000 0.0000 0.0000 -0.1765 -0.1765
2 0.6000 0.1985 0.0000 0.0000 0.0000 -0.2308 -0.2425
3 0.8250 0.6275 0.3333 0.4000 0.3636 0.2632 0.2646
4 0.7750 0.7353 0.3333 0.2857 0.3077 0.1743 0.1751
5 0.6750 0.5123 0.1667 0.1111 0.1333 -0.0569 -0.0587
6 0.7250 0.5779 0.1429 0.1667 0.1538 -0.0092 -0.0092
7 0.6000 0.3571 0.1429 0.0909 0.1111 -0.1307 -0.1363
8 0.6250 0.3658 0.0000 0.0000 0.0000 -0.2295 -0.2303
9 0.8500 0.4784 0.2857 0.6667 0.4000 0.3296 0.3685
Mean 0.7100 0.4765 0.1405 0.1721 0.1470 -0.0224 -0.0204
Std 0.0838 0.1441 0.1323 0.2093 0.1499 0.1964 0.2056
No description has been provided for this image
In [32]:
#  Finalize model (train on full dataset)
final_model = finalize_model(best_model)
In [33]:
#  Save model for deployment
save_model(final_model, "customer_delinquency_model")

Transformation Pipeline and Model Successfully Saved
Out[33]:
(Pipeline(memory=Memory(location=None),
          steps=[('numerical_imputer',
                  TransformerWrapper(exclude=None,
                                     include=['Age', 'Income', 'Credit_Score',
                                              'Credit_Utilization',
                                              'Missed_Payments', 'Loan_Balance',
                                              'Debt_to_Income_Ratio',
                                              'Account_Tenure'],
                                     transformer=SimpleImputer(add_indicator=False,
                                                               copy=True,
                                                               fill_value=None,
                                                               keep_empty_features=False,
                                                               missing_values=nan,
                                                               s...
                                     transformer=FixImbalancer(estimator=SMOTE(k_neighbors=5,
                                                                               random_state=None,
                                                                               sampling_strategy='auto')))),
                 ('normalize',
                  TransformerWrapper(exclude=None, include=None,
                                     transformer=StandardScaler(copy=True,
                                                                with_mean=True,
                                                                with_std=True))),
                 ('actual_estimator',
                  KNeighborsClassifier(algorithm='auto', leaf_size=30,
                                       metric='minkowski', metric_params=None,
                                       n_jobs=-1, n_neighbors=5, p=2,
                                       weights='uniform'))],
          verbose=False),
 'customer_delinquency_model.pkl')
In [34]:
#  Predict on new/unseen data
predictions = predict_model(final_model, data=df)
  Model Accuracy AUC Recall Prec. F1 Kappa MCC
0 K Neighbors Classifier 0.6140 0.9943 1.0000 0.2930 0.4533 0.2735 0.3980
In [35]:
# Show sample predictions
predictions.head()
Out[35]:
Customer_ID Age Income Credit_Score Credit_Utilization Missed_Payments Loan_Balance Debt_to_Income_Ratio Employment_Status Account_Tenure ... Month_4 Month_5 Month_6 DTI_bin Credit_bin Util_bin PayHist_bin Delinquent_Account prediction_label prediction_score
0 CUST0001 56 165580.0 398.0 0.390502 3 16310.0 0.317396 Employed 18 ... Late Missed Late (0.2, 0.4] (300, 500] (0.3, 0.5] (2.0, 5.0] 0 0 0.6
1 CUST0002 69 100999.0 493.0 0.312444 6 17401.0 0.196093 Self-employed 0 ... Missed On-time On-time (0.0, 0.2] (300, 500] (0.3, 0.5] (5.0, 10.0] 1 1 1.0
2 CUST0003 46 188416.0 500.0 0.359930 0 13761.0 0.301655 Self-employed 1 ... On-time Missed Late (0.2, 0.4] (300, 500] (0.3, 0.5] NaN 0 1 0.8
3 CUST0004 32 101672.0 413.0 0.371400 3 88778.0 0.264794 Unemployed 15 ... Missed Late Late (0.2, 0.4] (300, 500] (0.3, 0.5] (2.0, 5.0] 0 0 0.6
4 CUST0005 60 38524.0 487.0 0.234716 2 13316.0 0.510583 Self-employed 11 ... Late Late Late (0.4, 0.6] (300, 500] (0.0, 0.3] (0.0, 2.0] 0 0 0.8

5 rows × 25 columns

In [36]:
# Classification Report
plot_model(final_model, plot="class_report")
No description has been provided for this image

This KNN model prioritizes catching all delinquent customers (Recall = 100%), which is great for risk minimization.

But it sacrifices precision, meaning many safe customers get falsely flagged as delinquent.

In [37]:
#Probability Distribution
plot_model(final_model, plot="calibration")
No description has been provided for this image
In [38]:
# Sort customers by delinquency risk
risk_ranking = predictions.sort_values("prediction_score", ascending=False)
risk_ranking[["Customer_ID", "prediction_label", "prediction_score"]].head(10)
Out[38]:
Customer_ID prediction_label prediction_score
93 CUST0094 0 1.0
251 CUST0252 1 1.0
455 CUST0456 0 1.0
257 CUST0258 1 1.0
105 CUST0106 0 1.0
253 CUST0254 1 1.0
452 CUST0453 1 1.0
109 CUST0110 0 1.0
1 CUST0002 1 1.0
457 CUST0458 0 1.0
In [ ]:
In [ ]: